timeseries_markdown
Taylor Fergus
2024-05-24
First, exploring the best month to buy and sell a home in King County, Washington
I start by uploading the dataset and cleaning it.
library(readr)
kingcountysales <- read_csv("C:/Users/tjf4x/Desktop/R projects/King county home sales/kingcountysales_2000_2023.csv")
# Clean and transforming data
library(tidyverse)
library(lubridate)
glimpse(kingcountysales)## Rows: 575,319
## Columns: 49
## $ ...1 <dbl> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16…
## $ sale_id <chr> "1999..144", "1999..258", "1999..331", "1999..609", "…
## $ pinx <chr> "..2734100475", "..1535200725", "..6028000255", "..12…
## $ sale_date <chr> "1/5/1999", "1/5/1999", "1/4/1999", "1/11/1999", "1/7…
## $ sale_price <dbl> 150000, 235000, 293000, 178506, 270000, 184250, 17500…
## $ sale_nbr <dbl> 1, 1, 1, 1, 1, 1, 1, NA, 1, 1, 1, NA, 1, 1, 1, NA, 1,…
## $ sale_warning <chr> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, "34", NA,…
## $ join_status <chr> "demo", "demo", "demo", "demo", "demo", "demo", "demo…
## $ join_year <dbl> 1999, 1999, 1999, 1999, 1999, 1999, 1999, 1999, 1999,…
## $ latitude <dbl> 47.54436, 47.42125, 47.57210, 47.68557, 47.76448, 47.…
## $ longitude <dbl> -122.3208, -122.4381, -122.1223, -122.1857, -122.2081…
## $ area <dbl> 78, 100, 31, 74, 38, 8, 18, 49, 70, 81, 79, 55, 69, 2…
## $ city <chr> "SEATTLE", "KING COUNTY", "KING COUNTY", "KIRKLAND", …
## $ zoning <chr> "SF 5000", "RA2.5P", "R6", "RS8.5", "R15 OP", "SF 720…
## $ subdivision <chr> "GEORGETOWN", "CHAUTAUQUA BEACH ADD", "NELSONS H E EA…
## $ present_use <dbl> 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,…
## $ land_val <dbl> 21000, 46000, 66000, 80000, 73000, 84000, 35000, 9100…
## $ imp_val <dbl> 66000, 132000, 207000, 69000, 119000, 75000, 122000, …
## $ year_built <dbl> 1900, 1916, 1986, 1961, 1924, 1920, 1955, 1986, 1967,…
## $ year_reno <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ sqft_lot <dbl> 4000, 6695, 12360, 9450, 13600, 7200, 12554, 20300, 1…
## $ sqft <dbl> 1410, 990, 2020, 880, 2050, 790, 2160, 2230, 560, 151…
## $ sqft_1 <dbl> 760, 990, 1470, 880, 1300, 790, 1440, 1240, 560, 750,…
## $ sqft_fbsmt <dbl> 0, 0, 0, 0, 0, 0, 720, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ grade <dbl> 7, 6, 8, 6, 7, 7, 6, 9, 3, 7, 6, 5, 7, 8, 7, 7, 7, 6,…
## $ fbsmt_grade <dbl> 0, 0, 0, 0, 0, 0, 6, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6,…
## $ condition <dbl> 3, 4, 3, 3, 5, 3, 3, 3, 3, 5, 3, 4, 3, 4, 5, 3, 3, 3,…
## $ stories <dbl> 1.5, 1.0, 2.0, 1.0, 1.5, 1.0, 1.0, 1.5, 1.0, 2.0, 1.0…
## $ beds <dbl> 3, 2, 3, 2, 5, 2, 4, 4, 1, 3, 3, 4, 4, 3, 3, 3, 3, 3,…
## $ bath_full <dbl> 1, 1, 1, 1, 1, 1, 1, 2, 0, 1, 1, 1, 1, 1, 2, 1, 1, 1,…
## $ bath_3qtr <dbl> 0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1,…
## $ bath_half <dbl> 0, 0, 1, 0, 1, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0,…
## $ garb_sqft <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ gara_sqft <dbl> 0, 0, 480, 0, 0, 0, 0, 730, 0, 0, 0, 0, 480, 580, 0, …
## $ wfnt <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 9, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ golf <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ greenbelt <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ noise_traffic <dbl> 2, 0, 2, 1, 2, 1, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 2, 0,…
## $ view_rainier <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ view_olympics <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ view_cascades <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ view_territorial <dbl> 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ view_skyline <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ view_sound <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ view_lakewash <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ view_lakesamm <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ view_otherwater <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ view_other <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ submarket <chr> "E", "H", "R", "Q", "Q", "C", "F", "G", "N", "E", "E"…# Removing duplicates
distinct(kingcountysales)## # A tibble: 575,319 × 49
## ...1 sale_id pinx sale_date sale_price sale_nbr sale_warning join_status
## <dbl> <chr> <chr> <chr> <dbl> <dbl> <chr> <chr>
## 1 1 1999..144 ..27… 1/5/1999 150000 1 <NA> demo
## 2 2 1999..258 ..15… 1/5/1999 235000 1 <NA> demo
## 3 3 1999..331 ..60… 1/4/1999 293000 1 <NA> demo
## 4 4 1999..609 ..12… 1/11/1999 178506 1 <NA> demo
## 5 5 1999..775 ..19… 1/7/1999 270000 1 <NA> demo
## 6 6 1999..957 ..52… 1/15/1999 184250 1 <NA> demo
## 7 7 1999..1155 ..34… 1/19/1999 175000 1 <NA> demo
## 8 8 1999..1267 ..06… 1/11/1999 359850 NA <NA> demo
## 9 9 1999..1335 ..73… 1/7/1999 105000 1 <NA> demo
## 10 10 1999..1354 ..33… 1/14/1999 160000 1 <NA> demo
## # ℹ 575,309 more rows
## # ℹ 41 more variables: join_year <dbl>, latitude <dbl>, longitude <dbl>,
## # area <dbl>, city <chr>, zoning <chr>, subdivision <chr>, present_use <dbl>,
## # land_val <dbl>, imp_val <dbl>, year_built <dbl>, year_reno <dbl>,
## # sqft_lot <dbl>, sqft <dbl>, sqft_1 <dbl>, sqft_fbsmt <dbl>, grade <dbl>,
## # fbsmt_grade <dbl>, condition <dbl>, stories <dbl>, beds <dbl>,
## # bath_full <dbl>, bath_3qtr <dbl>, bath_half <dbl>, garb_sqft <dbl>, …Selecting appropriate variables and converting variables.
I converted the cities into factors in case I wanted to do further analysis broken down by city.
# selecting variables and converting date
king_time <- kingcountysales %>%
select(sale_date, sale_price, city) %>%
mutate(date = mdy(sale_date))
# converting cities to factors
unique(king_time$city)## [1] "SEATTLE" "KING COUNTY" "KIRKLAND" "BOTHELL"
## [5] "NORMANDY PARK" "PACIFIC" "ISSAQUAH" "COVINGTON"
## [9] "KENT" "BELLEVUE" "SHORELINE" "FEDERAL WAY"
## [13] "REDMOND" "BURIEN" "MERCER ISLAND" "LAKE FOREST PARK"
## [17] "KENMORE" "AUBURN" "HUNTS POINT" "TUKWILA"
## [21] "RENTON" "MAPLE VALLEY" "SEA-TAC" "CLYDE HILL"
## [25] "MEDINA" "SNOQUALMIE" "ENUMCLAW" "NEWCASTLE"
## [29] "DES MOINES" "NORTH BEND" "BLACK DIAMOND" "WOODINVILLE"
## [33] "DUVALL" "CARNATION" "ALGONA" "YARROW POINT"
## [37] "BEAUX ARTS" "SKYKOMISH" "MILTON" "SAMMAMISH"
## [41] "SeaTac"king_time$city <- as.factor(king_time$city)
# Checking for missing values
sum(is.na(king_time))## [1] 0Calculating the monthly median home sale and creating a series of plots.
# Calculating median by month
library(timetk)
king_monthly <-
king_time %>%
summarize_by_time(.date_var = date,
.by = "month",
median_sale = median(sale_price, na.rm = TRUE))
head(king_monthly, 12)## # A tibble: 12 × 2
## date median_sale
## <date> <dbl>
## 1 1999-01-01 222000
## 2 1999-02-01 225475
## 3 1999-03-01 223250
## 4 1999-04-01 230000
## 5 1999-05-01 226000
## 6 1999-06-01 235000
## 7 1999-07-01 235000
## 8 1999-08-01 235920
## 9 1999-09-01 235000
## 10 1999-10-01 229970
## 11 1999-11-01 236250
## 12 1999-12-01 233800# Plotting monthly median sale price
plot_time_series(king_monthly,
.date_var = date,
.value = median_sale,
.interactive = TRUE,
.x_lab = "Monthy Data",
.y_lab = "Median Sale price")# Plotting monthly sales to find best months to buy or sell homes in king county
plot_seasonal_diagnostics(king_monthly, .date_var = date, .value = median_sale)# Best month to sell by median home price is June
# Best month to buy is JanuaryBecause more recent trends may effect home sale prices after the housing crisis I removed sales sales before January 2013 to see if there was a change.
# Removing data near housing crisis to see if it changes monthly medians
king_monthly_post2012 <-
king_monthly %>%
filter_by_time(.date_var = date,
.start_date = "2013",
.end_date = "2024")
glimpse(king_monthly_post2012, 12)## Rows: 132
## Columns: 2
## $ date <date> …
## $ median_sale <dbl> …plot_seasonal_diagnostics(king_monthly_post2012, .date_var = date, .value = median_sale)# This has similar results with June still being the best month to sell
# but December edging out January as the best month to buy.I wanted to see the accuracy of a simple univariant forecast model of the median home price. I did this using the Fable package and compared a couple different options
# Building a simple univariant forecast model
# -------------------------------------------
# Creating testing and training split
library(rsample)
king_monthly_split <- initial_time_split(king_monthly_post2012, prop = 130/136)
king_training <- training(king_monthly_split)
king_testing <- testing(king_monthly_split)
# converting data to a tibble and setting index as date
library(fable)
library(tsibble)
king_training <-
king_training %>%
mutate(date = yearmonth(date)) %>%
as_tsibble(index = date)
king_testing <-
king_testing %>%
mutate(date = yearmonth(date)) %>%
as_tsibble(index = date)
king_monthly_post2012 <-
king_monthly_post2012 %>%
mutate(date = yearmonth(date)) %>%
as_tsibble(index = date)
# Train the models
library(feasts)
king_fit <-
king_training %>%
model(stepwise = ARIMA(median_sale),
search = ARIMA(median_sale, stepwise=FALSE))
# Viewing model
tidy(king_fit)## # A tibble: 6 × 6
## .model term estimate std.error statistic p.value
## <chr> <chr> <dbl> <dbl> <dbl> <dbl>
## 1 stepwise ar1 0.857 0.0488 17.5 3.70e-34
## 2 stepwise sma1 -0.789 0.0945 -8.35 1.82e-13
## 3 stepwise constant 7027. 584. 12.0 4.84e-22
## 4 search ar1 0.857 0.0488 17.5 3.70e-34
## 5 search sma1 -0.789 0.0945 -8.35 1.82e-13
## 6 search constant 7027. 584. 12.0 4.84e-22king_fit %>%
accuracy() %>%
arrange(MAPE)## # A tibble: 2 × 10
## .model .type ME RMSE MAE MPE MAPE MASE RMSSE ACF1
## <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 stepwise Training -426. 18360. 13356. -0.188 1.98 0.251 0.283 0.0519
## 2 search Training -426. 18360. 13356. -0.188 1.98 0.251 0.283 0.0519# Checking forecast of training data vs full data set
king_fit %>%
forecast(h = "6 months") %>%
autoplot(king_monthly_post2012)
The model performed very well. Better than I expected given it is only using past median sales data without any external economic factors
I decided to train the model on the full dataset that ends in December of 2023 and see how it performs compared to current monthly median sales.
# final fit on full data set
king_final_fit <-
king_monthly_post2012 %>%
model(stepwise = ARIMA(median_sale),
search = ARIMA(median_sale, stepwise=FALSE))
# Fitting final model on full data set
king_final_fit %>%
forecast(h = "12 months") %>%
autoplot(king_monthly_post2012) +
labs(x = "Month", y = "median home price")
king_final_fit %>%
forecast(h = "12 months") %>%
autoplot() +
labs(x = "Month", y = "median home price")
king_monthly_t <-
king_monthly %>%
mutate(date = yearmonth(date)) %>%
as_tsibble(index = date)Compared to the latest data for April of 2024, the king county median home sale was $895,000. That is still in the 80% confidence interval but right on the low end. The model is being too optimistic. This is likely because the model has mostly only seen data since the recovery from the 2008 financial crisis.
I added the full data set back in (since 2002) and retrained the model.
# The data ends in December of 2023, but referencing median sales in recent
# we can see the forecast starts to get too optimistic. I added back in the housing
# crisis data and retrained.
king_final_fit_full <-
king_monthly_t %>%
model(stepwise = ARIMA(median_sale),
search = ARIMA(median_sale, stepwise=FALSE))
king_final_fit_full %>%
forecast(h = "12 months") %>%
autoplot(king_monthly_t) +
labs(x = "Month", y = "median home price")
king_final_fit_full %>%
forecast(h = "12 months") %>%
autoplot() +
labs(x = "Month", y = "median home price")
# Adding the full time series makes the model more conservative and makes the
# forecast more accurate.